Hello! Please tell me, can I train a model on data from articles on my website, so that in the future chatGPT will respond only based on this data?
but I wouldn’t call it training - industry standard at the moment is to use Retrieval Augmented Generation (RAG) to accomplish the task you’re describing.
You would use the embedding models (like embed ada 002) to build an index of your website’s content, push it into a vector database, and then retrieve (with ada again) it and push it into your chatbot’s context to allow it to formulate a relevant answer.
and welcome to the developer community forum!
Almost, you can’t change how ChatGPT functions, that’s an OpenAI product. But you can create your own fine tune for using the API
There’s different approaches to this depending on what you like to achieve, are you interested in having GPT help you write content for your website, or do you want GPT to answer questions about the content on your site?
My goal is to ensure that GPT can provide short answer to user’s question based on information from articles
Then this is what you want:
There’s an example in the OpenAI cookbook here:
You can use fine tuning on top of that if you want to improve the results even further, here’s an example of that:
I hope that helps
let us know if you have any conceptual questions with your pursuit